Research And Implementation Of Face Recognition Based On Embedded System

With the rapid development of embedded system, face recognition based onembedded system has been a hot research subject in the fields ofpattern recognition and artificial intelligence, which has great application valuesin public security, e-commerce, administration of justice, finance and so on.The work of this thesis is supported by the Industry, Education and ResearchProject of Guangdong Province (No.20100901). The target of this project is tocomplete a face recognition component on embedded system with key specifications ofrecognition rate of85%and recognition speed of2frames per second. This thesisdiscussed several key issues such as face detection, face pre-processing andface recognition. The contribution and innovation can be summarized as follows.1、 This thesis researched original AdaBoost based face detection algorithm,explained the main reason of its low speed, and presented a new way, which reduces theaccuracy but increases the speed of face detection. The methods of optimization are:changing the step of the Haar Sub-window, changing of the detection order, convertingfloat point operations into fixed point operations, cutting the Cascade classifier,detecting the scaling images. The optimized method has over2.5times more rapid thanoriginal algorithm.2、This thesis presented two rapid eye localization algorithm：eye location basedon AdaBoost and based on eyeballs quick search. The premise of both methods is thatthe face areas has been detected, and then detect two eyes in further narrowed detectionarea. The experiment proved that the two methods are fast and efficient.3、This thesis presented improved cascade LBP-based face recognition algorithmbased on original LBP-based face recognition algorithm. This algorithmuses “non-uniform patterns” effectively while making full use of the "uniform patterns",and the histogram Features can represent the face effectively. Finally, wemade experiment on ORL face database and FERET standard face database using theimproved algorithm and compared to with several other face recognition algorithms,proved the effectiveness of the proposed algorithm. 4、Finally, this thesis introduced the implementation of face recognition system onTiny210development board based on ARM architecture processor andfinished two applications: FacePlayer and FaceCar which are music player and toy carcontroller based on face recognition. The face recognition system has the recognitionrate of85.2%and the recognition speed of more than2frames per second after makingexperiment on images gathered from actual application environment.